On 8/25/15, Joseph Wright <[log in to unmask]> wrote:
> On 25/08/2015 06:16, Will Robertson wrote:
>> On 26 Aug 2015, at 09:14, aparsloe <[log in to unmask]> wrote:
>>>
>>> (1) I wanted to use \tl_replace_all:Nnn on a token list that might
>>> contain braced groups.
>>
>> We might want to consider a "nesting" map function that essentially
>> disregards brace groups. This could be used to define a new
>> \tl_replace_nested:Nnn or similar for exactly this purpose.
>>
>> I'm fairly sure Bruno already has code that does nested mapping as part of
>> his "tl" extensions... What do others think?
>
> I guess I'd want to know a bit more about use cases. As I've said on
> LaTeX-L, normally brace groups are an 'escape' mechanism from the
> scanner, so I'm not sure I'd expect recursive replacement as a general
> thing. I'm wondering if this is being used for something rather 'text
> like', which might be handled by l3regex (slow but for a smaller number
> of cases reasonable).

I'm not completely sure that l3regex will be slower in all cases than
an approach based on my nested mappings.  Besides, if we find some
common use-cases of regexes that can be optimized, I can work on that.
For instance, I saw someone use \c{alpha}|\c{beta}|..., which is
currently quite slow but can be optimized.

For your information, "aparsloe", l3regex provides \u, to use the
contents of a tl variable in the search or the replace part of a
regular expression.

Bruno